JP2010504662A - Maximum logarithmic receiver for multiple input multiple output (MIMO) systems - Google Patents

Abstract

The method i) lists all candidates for the first symbol of the two stream signals, and ii) determines the second symbol of the two stream signals for each of the listed first symbols. Iii) evaluate the metric for each of the first symbol and second symbol pair, iv) list all candidates for the second symbol, and v) the second symbol listed For each selection from, vi) evaluate the metric for each of the second symbol and first symbol pair, and viii) the exact maximum log likelihood determined for all bits Decoding the codewords in the two stream signals using the degree ratio.
[Selection] Figure 1

Description

  This application claims the benefit of US Provisional Application No. 60 / 826,119, filed September 19, 2006, entitled “New Two-Stream Receiver and Extension for MIMO Systems” The contents are incorporated herein by reference.

  The present invention relates generally to mobile communications, and more particularly to a two-stream receiver for a multiple-input multiple-output MIMO system and its extensions.

  To meet the ambitious spectral efficiency goals set for Advanced UMTS Terrestrial Radio Access (EUTRA), low latency, low complexity receivers are needed. Such a receiver is particularly needed in user equipment (UE) where the complexity constraints are very strict. The most important scenario in a multi-antenna downlink system involves a UE with two antennas, and the base station or Node-B transmits two encoded streams to the scheduled UE.

  The known aggressive maximum likelihood ML reception method 10 for the two streams shown in FIG. 1 lists all possible pairs 11 for symbols 1 and 2, 11 and evaluates the metric for each of the pairs 12, using that metric to determine the exact maximum logarithm LLR (maximum log likelihood ratio) for all bits 13, and using that calculated LLR to decode the codeword 14 to do. The aggressive ML method provides optimal demodulation, but it is very complex.

  The main demodulator competing for the present invention is the deterministic sequential Monte Carlo (D-SMC) based receiver (another promising low complexity receiver) shown in FIG. 2 and FIG. The continuous interference cancellation SIC receiver shown in FIG.

  Complexity reduction is achieved with the D-SMC method by calculating the soft output for each bit coded from only the reduced hypothesis group. The price paid for this complexity reduction is usually that the hypotheses (or candidates) needed to compute the soft output for some bits may not exist in the reduced group. D-SMC is affected by the problem, which is called “missing candidate problem”. This missing candidate problem can cause significant performance degradation, especially when the reduced group is relatively small compared to all hypothetical groups. Heuristics have also been proposed to alleviate this problem in D-SMC, but such techniques require specific fine tuning of many systems or scenarios and may not work well under all conditions .

  Referring back to FIG. 2, the D-SMC method 20 lists 21 a subset of possible pairs of symbols 1 and 2 for the two received streams 21 and evaluates the metric for each of the pairs. 22, using the metric to determine an approximate maximum logarithm LLR for all bits 23, and decoding the codeword using the calculated LLR 24. The D-SMC reception method has adjustable complexity, but it also has sub-optimal demodulation due to the “missing” candidate problem.

  In contrast to the D-SMC reception method, the SIC receiver is a sequential receiver, in which one stream is first decoded and the received signal before decoding the second stream. Deducted from. The soft output for the first stream is obtained after estimating the second stream, which is a Gaussian interferer that causes performance degradation.

  Referring again to FIG. 3, the continuous interference cancellation reception method 30 uses the filter to suppress the contribution of symbol-2 31, evaluate the metric 32 for all candidates of symbol-1, and use that metric. Determining, for all bits associated with symbol-1, 33 the maximum logarithm LLR, decoding the codeword-1 using the calculated LLR 34, and then modulating and receiving from the received signal Re-encode the subtracted codeword 35, list all candidates for symbol-2, and calculate its metric 36, using that metric, all related to symbol-2 Determining 37 the maximum logarithm LLR for the bits of, and then decoding 38 codeword-2

  Thus, for a two-stream receiver that is highly suitable for a receiver with the low latency and low complexity required to meet the ambitious spectral efficiency goals set for Advanced UMTS Terrestrial Radio Access (EUTRA) There is a need.

  According to the invention, the method i) lists all candidates for the first symbol of two stream signals, and ii) for each of the listed first symbols, Determine a second symbol, iii) evaluate a metric for each of the first symbol and second symbol pair, iv) list all candidates for the second symbol, and v) a list For each selection from the second symbol up, determine a first symbol, vi) evaluate a metric for each of the second symbol and first symbol pair, and vii) use the metric Determine the exact maximum log-likelihood ratio for all bits, and viii) the exact maximum determined for all bits Using a logarithmic likelihood ratio to decode the codewords in said two stream signal, comprising the steps.

  In another aspect of the invention, the method performs a CRC check on the two decoded codewords, i) to viii) for decoding two codewords associated with two streams, and ix) x) If the CRC of only one codeword is true, the codeword is re-encoded, modulated and subtracted from the received signal to obtain one stream signal, xi) within the one stream signal List all candidates for the remaining symbols, xii) evaluate the metric for each candidate of the remaining symbols, and xiii) use the metric to determine the maximum log likelihood ratio for all bits And xiv) decoding the remaining codewords in the one stream signal using the maximum log likelihood ratio determined for all bits, Tsu, including the flop.

  In a preferred extension, the step of receiving two signal streams is extended to four received signal streams by dividing the demodulation of the four signal streams into two smaller two stream signal demodulations. The demodulation of the two, smaller, two stream signals can be resolved sequentially, such as sequential group decoding, or in parallel, such as parallel group decoding. The parallel group decoding consists of four received streams (labeled {1, 2, 3, 4}), {(1,2), (3,4)}, {(1,3 ), (2, 4)}, {(1, 4), (2, 3)} including splitting into one of the three disordered sections {1, 2, 3, 4} Yes. The segmentation is performed when the four streams are coded jointly, as in the case of a single codeword (SCW), or as if the four streams are in multiple codewords (MCW). In addition, it can be done on a per-tone basis (in the OFDM scheme with multiple tones) based on the instantaneous channel realization, taking into account the case of independent coding. The decoding of sequential groups is {(1,2), (3,4)}, {(3,4), (1,2)}, {(1,3), (2,4)}, { {1,2,3,4, {2,4), (1,3)}, {(1,4), (2,3)}, {(2,3), (1,4)} } In order to allow for post-decoding feedback and independently using a common or fixed partition across all tones Requires a stream to be converted.

In the context of the present invention, consider the joint demodulation of two streams, where each stream is composed of symbols from a constellation of size M. Instead of the traditional O (M ² ) complexity method of evaluating all M ² hypothesis metrics, by evaluating the 2M hypothesis metric, for every 2 log (M) bits per symbol interval, An accurate maximum logarithmic output is obtained using O (M) complexity. Based on this, an inventive maximum log 2 stream receiver, which is the flow diagrammed in FIG. 4, is provided.

  In another aspect of the invention, a two-stream extended maximum log receiver is provided, where the maximum log receiver is first used to decode two codewords. If only one codeword is correctly decoded, the correctly decoded codeword is re-encoded, modulated and subtracted from the received signal. The remaining codewords (incorrectly decoded in the first attempt) are decoded again using the signal thus obtained. An inventive extended maximum logarithmic receiver is shown in the flow chart of FIG.

  A method for extending the inventive two-stream receiver to multiple streams is also described, with particular emphasis on the case of four streams, another important scenario.

  Referring back to FIG. 4, the inventive maximum log reception method 40 lists 41 all candidates for symbol-1, 41, for each selection from symbol-1, the best choice of symbol-2 is efficient. Find and evaluate the metric for the pair 42, list all candidates for symbol-2 43, efficiently find the best choice for symbol-1 for each choice from symbol-2 , Evaluating the metric for the pair 44, using the metric to determine the exact maximum logarithm LLR for all bits 45, and using the calculated LLR, the received codeword Is included. This inventive maximum log reception method provides optimal demodulation with low execution complexity.

  We immediately describe an inventive two-stream maximum log demodulator. Consider the following model:

Here, H is an N × 2 channel matrix (N ≧ 2), and ν is additive noise having a Gaussian element with i.i.d. zero mean unit variance.

x ₁ and x ₂ are symbols from a common M-QAM constellation. H = [h ₁ , h ₂ ], and H = ‖h ₂ ‖ ² UL, where U is a scaled semi-unitary matrix, and L is a lower triangle with positive diagonal elements. Is the modified QR decomposition.

  In particular, we

To obtain U = [u ₁ , u ₂ ].
Where <h ₁ , h ₂ > = h ₂ ^* h ₁ is the (complex) inner product of two vectors, and

Next, we

Get the transformed noise vector

Note that remains white.

Represents an M-QAM symbol, and x _i ^R and x _i ^I represent a real part and an imaginary part of x _i (where 1 ≦ i ≦ 2). For each x1 _{, j} we define the following metric:

If we define q _1j = z ₂ −L ₂₁ x _{1, j} , we can express Q (x _{1, j} ) as:

x ₂ ^R and x ₂ ^I are both common

Since it belongs to the constellation, the two minimizations for computing Q (x _{1, j} ) are in parallel using simple slicing (rounding) operations with O (1) complexity respectively. Can be done.

All {Q (x1 _{, j} )} _{j = 1} ^M is efficiently determined using the above method. By also using the fact that L ₁₁ is positive with the symmetry of the M-QAM constellation, we

Get. next,

So we all

To evaluate

It can be seen that only (real) multiplication is required instead of 2M imaginary multiplication.

Then we obtain the QR decomposition H = ‖h ₁ ‖ ² VR which is another modification, however, V is, it scaled Semiyunitari matrix, and, R represents at upper triangular with positive diagonal elements is there.

In particular, we obtain V = [ν ₁ , ν ₂ ], where

Here, <h ₂ , h ₁ > = h ₁ ^* h ₂ is a complex conjugate of <h ₁ , h ₂ >

It is.

Using V, we determine w = V ^* y, which can be expressed as:

Next, for each x2 _{, j} we define the following metric.

By defining q _2j = w ₁ −R ₁₂ x _{2, j} , we can express Q (x _{2, j} ) as

Again, x ₁ ^R and x ₁ ^I are both

Since it belongs to the constellation, the two minimizations for calculating Q (x _{2, j} ) can also be performed in parallel using simple slice operations as described above. All {Q (x _{2, j} )} _{j = 1} ^M is efficiently determined using the above method. 2M metrics

However, it can be determined efficiently even for other fixed constellations. For example, consider an example of a PSK constellation. Let x ₁ and x _{2 be} the average energy M-PSK constellation of the common unit

Symbol from Next, in order to efficiently determine {Q (x1 _{, j} )}, we rewrite equation (2b) as

Then we obtain q _{1, j} in its polar form, such as q _{1, j} = r _{1, j} exp (jα _{1, j} ), where r _{1, j} > 0,

And Let β1 _{, j} = Mα1 _{, j} / (2π) −1/2.

Next, where the minimization x ₂ of the formula (3b) is a closed form (with the complexity of O (1)) can be determined,

Given here, where

Indicates a floor operator. Similarly, we can efficiently determine the minimization x ₁ in equation (2c) using the complexity of O (1). In a similar manner, the minimization x ₁ in equation (2c) (and the minimization x ₂ in equation (2b)) is efficient for other fixed constellations by exploiting their decision regions. Can be determined.

  Here, since each size M constellation corresponds to log (M) bits, we need to determine the maximum logarithmic soft output for 2 log (M) bits. 2M metrics we have determined efficiently

Is exactly what is needed to determine the maximum logarithmic output for each of the bits. To understand this, assume that the bit numbered 1 in log (M) corresponds to the symbol x ₁ . Next, assuming that λ _i represents the maximum logarithmic output of the i th bit b _i and that the a priori probabilities are equal, we have

and

Get.
Thus, we see that the complexity of our method for determining the maximum logarithmic output for each of the 2 log (M) bits is O (M), not the O (M ² ) complexity of the normal method Is shown. Note that if the two symbols belong to different constellations, the method is extended to the case in a straightforward manner.

  Further reduction in complexity is due to two modified QR decompositions

Can be achieved by avoiding redundant computations. A significant reduction in processing delays

and

Can be achieved by performing the computations in parallel.

  The inventive maximum log 2 stream receiver includes the 2 stream demodulator described above along with an outer code (FEC) encoder.

  Referring again to FIG. 5, the inventive extended maximum log receiver 50 method for receiving two streams decodes two codewords using the maximum log receiver and the two decoded codes. Perform cyclic redundancy check (CRC) 51 on word, check if CRC is true for both codewords or CRC is false for both codewords If steps 52, 52 are true If the steps 53 and 52 are false, check if the CRC of codeword-1 is true. If the steps 54 and 54 are true, codeword-1 And then modulating and subtracting the codeword-1 from the received signal 55, listing all candidates for symbol-2 and calculating their metric 56, using the metric , Symbol-2 Determining the maximum logarithm LLR 57 for all the consecutive bits and decoding the codeword-2 using the calculated LLR 58, if the steps of 54, 54 are false, the codeword − Re-encode 2 and then modulate and subtract codeword-2 from the received signal 59, list all candidates for symbol-1 and calculate its metric 60, using the metric, Determining 61 a maximum logarithmic LLR for all bits associated with symbol-1 and decoding 62 codeword-1 using the calculated LLR. This inventive extended maximum log receiver method has higher complexity and latency (delay), and a need for more memory for buffering, but performs better than the maximum log receiver. It has been improved.

  Next, we describe our extended maximum log receiver. Our extended maximum log receiver operates as follows. We decode two codewords using the maximum log receiver described above and perform a cyclic redundancy check (CRC) on the two decoded codewords. If the CRC is true for both codewords or false for both codewords, we stop the encoding process. If the CRC is true for codeword-1 (false for codeword-2) every symbol interval,

Where

Corresponds to codeword-1 which has been re-encoded and modulated, and the soft output for the second stream (codeword) is obtained as:

The resulting LLR is used to decode the second codeword.

  If the CRC is true for codeword-2 (false for codeword-1) at every symbol interval, we

And the soft output for the first stream (codeword) is obtained as:

The resulting LLR is used to decode the first codeword.

  In order to extend our maximum log 2 stream receiver and decode many streams, we use the group decoding concept. The resulting receiver no longer produces an accurate maximum logarithmic output for each of the coded bits, but nevertheless the receiver provides excellent performance with low complexity. . For example, consider the case of 4-stream transmission over MIMO-OFDM. On each of the N tones, we have a flat fading MIMO model given by

  By dividing the 4-stream demodulation problem into two smaller 2-stream demodulation problems that are subsequently solved by our 2-stream demodulator, we can take advantage of our 2-stream demodulator. Furthermore, the two smaller problems can be solved sequentially (like continuous group coding) or in parallel (like parallel group coding).

  In the parallel case we are {(1,2), (3,4)}, {(1,3), (2,4)}, {(1,4), (2,3)} There are three methods for performing the division corresponding to the three disordered sections {1, 2, 3, 4}. This splitting is achieved when the four streams are coded jointly, as in the case of a single codeword (SCW), or the four streams are independent as in the case of multiple codewords (MCW). Can be done on a per-tone basis based on the instantaneous channel realization. Note that in the case of SCW, only the maximum log demodulator can be used for smaller two-stream problems. To elaborate further, assume that {(1,2), (3,4)} is a chosen division on several tones. Next, in parallel group decoding, we suppress the streams 3 and 4 using MMSE filtering, whiten the noise and the suppressed interference, and then use a two-stream demodulator to stream Obtain the LLR for 1 and 2. Similarly, we use LMM for streams 3 and 4 by using MMSE filtering to suppress streams 1 and 2 and whitening the noise and the suppressed interference, then using a 2-stream demodulator. Get.

  In sequential order, we have {(1,2), (3,4)}, {(3,4), (1,2)}, {(1,3), (2,4)}, { {1,2,3,4, {2,4), (1,3)}, {(1,4), (2,3)}, {(2,3), (1,4)} } Has six ways to perform the partitioning corresponding to the six orderly sections. In this case, however, we need an independently encoded stream and the split should be shared or fixed across all tones to allow post-decoding feedback. is there. We can decode two codewords in each of two smaller two-stream problems using either one of two two-stream receivers. To further elaborate, assume that {(1,2), (3,4)} is the partition chosen across all tones. Next, in continuous group decoding, we use MMSE filtering to suppress streams 3 and 4, and after whitening noise and the suppressed interference, by using a two stream receiver, Streams 1 and 2 are decoded. Next, we decode streams 3 and 4 by subtracting the reconstructed streams 1 and 2 from the received signal and performing a complete removal of streams 1 and 2, using a two stream receiver To do.

  Next, if limited feedback is available, the receiver can choose one tone from three chaotic segments per tone, or six orderly segments (which are fixed across all tones). And can notify the transmitter. The transmitter can then use one codeword in each group, and continuous group decoding (using the maximum logarithmic demodulator for each group) can be used at the receiver. .

  In summary, we designed two receivers considering the two-stream MIMO decoding problem. The first receiver is a very efficient implementation of a maximum likelihood demodulator (MLD) that yields an accurate maximum log LLR output. The second receiver is an enhanced maximum log receiver that provides further performance improvements at the expense of higher complexity. An extension of the inventive two-stream receiver to the general case using multiple streams was also obtained.

  The present invention has been illustrated and described in what is considered to be the most practical and preferred embodiment. However, it is anticipated that new attempts may be made therefrom and obvious changes will be made by those skilled in the art. Those skilled in the art may devise many devices and variations that are not explicitly shown or described herein, but which embody the principles of the invention and are within the spirit and scope of the invention. It is expected that.

These and other advantages of the invention will be apparent to those of ordinary skill in the art by reference to the following detailed description and the accompanying drawings.
It is a flowchart of the aggressive maximum likelihood 2 stream radio | wireless reception which concerns on a prior art. 3 is a flowchart of deterministic sequential Monte Carlo (D-SMC) 2 stream reception according to the prior art. 3 is a flowchart of continuous interference cancellation (SIC) two-stream radio reception according to the prior art. 4 is a flowchart of maximum log 2 stream wireless reception according to the present invention. 4 is a flowchart of an extended maximum logarithmic receiver for two-stream wireless reception according to the present invention.

Claims (14)

i) list all candidates for the first symbol of the two stream signals;
ii) for each of the listed first symbols, determining a second symbol of the two stream signals;
iii) evaluating a metric for each of the first symbol and second symbol pairs;
iv) list all candidates for the second symbol,
v) For each selection from the listed second symbol, determine a first symbol;
vi) evaluating a metric for each of the second symbol and first symbol pair;
vii) using the metric to determine the exact maximum log-likelihood ratio for all bits;
viii) decoding codewords in the two stream signals using the exact maximum log-likelihood ratio determined for all the bits;
A method comprising steps.   The method of claim 1, wherein the first symbol is designated as symbol-1 and the second symbol is designated as symbol-2.   When symbol-1 and symbol-2 belong to a constellation of size M, the maximum log likelihood ratio for all 2 Log (M) bits in the two stream signals is obtained by O (M) complexity. The method of claim 1. The further reduction in complexity can be achieved by avoiding redundant computations in two modified QR decompositions, H = ‖h ₂ ‖ ² UL, H = ‖h ₁ ‖ ² VR. 3. The method according to 3. The reduction in processing delay is

and

The method of claim 4, wherein the calculation of can be accomplished by performing in parallel. The metric is

Defined in

Define

But both are common

Q (x1 _{, j} ) as belonging to the constellation

2 wherein the two minimizations for calculating Q (x1 _{, j} ) are performed in parallel using each slicing (rounding) operation using O (1) complexity. The method described in 1. The metric is

Based on

Define

But both

Q (x _{2, j} ) as belonging to the constellation

Expressed as, the two minimizations to calculate Q (x2 _{, j} ) are done in parallel using a simple slice operation, which

The method of claim 1, wherein the method is determined efficiently. If each of symbol-1 and symbol-2 is different, as in PSK, but belongs to a fixed constellation, the metric {Q (x1 _{, j} ), Q (x2 _{, j} )} is the efficiency The method of claims 6 and 7, wherein the method is determined automatically. 7. An efficiently determined metric {Q (x1 _{, j} ), Q (x2 _{, j} )} is used to determine the maximum log likelihood ratio LLR output for each bit. , 7 and 8. i) decoding two codewords associated with the two streams using the method of claim 1;
ii) perform a CRC check on the two decoded codewords;
iii) If the CRC of only one codeword is true, the codeword is re-encoded, modulated and subtracted from the received signal to obtain a single stream signal,
iv) list all candidates for the remaining symbols in the one stream signal;
v) For each remaining symbol candidate, evaluate the metric,
vi) Determine the maximum log likelihood ratio for all bits using the metric,
vii) decoding the remaining codewords in the one stream signal using the maximum log likelihood ratio determined for all bits.   11. The two signal stream reception steps are extended to four received signal streams by dividing the four signal streams into two smaller two stream signal demodulations. The method described in 1.   12. Demodulation of the two, smaller, two stream signals can be resolved sequentially, such as sequential group decoding, or in parallel, such as parallel group decoding. the method of.   The decoding of the parallel groups is {(1,2), (3,4)}, {(1,3), (2,4)}, {(1,4), (2,3)} and Comprising a split of the four incoming streams corresponding to one of the three chaotic segments {1, 2, 3, 4}, the four streams being a single codeword (SCW) Taking into account the case where they are coded jointly as in the case of, or where the four streams are coded independently as in the case of multiple codewords (MCW), 13. The method according to claim 12, which is performed on a per-tone basis (in the OFDM scheme used).   The sequential solutions are {(1,2), (3,4)}, {(3,4), (1,2)}, {(1,3), (2,4)}, {( 2,4), (1,3)}, {(1,4), (2,3)}, {(2,3), (1,4)} {1, 2, 3, 4} Including six methods for performing partitioning, the streams need to be encoded independently, and in order to allow post-decoding feedback, the partition is 13. The method of claim 12, wherein the method is common or fixed across all tones.

Images